In this study, we chose 17 worldwide sheep populations of eight breeds, which were intensively selected for different purposes (meat, milk, or wool), or locally-adapted breeds, in order to identify and characterize factors impacting the detection of runs of homozygosity (ROH) and heterozygosity-rich regions (HRRs) in sheep. We also applied a business intelligence (BI) tool to integrate and visualize outputs from complementary analyses. We observed a prevalence of short ROH, and a clear distinction between the ROH profiles across populations. The visualizations showed a fragmentation of medium and long ROH segments. Furthermore, we tested different scenarios for the detection of HRR and evaluated the impact of the detection parameters used. Our findings suggest that HRRs are small and frequent in the sheep genome; however, further studies with higher density SNP chips and different detection methods are suggested for future research. We also defined ROH and HRR islands and identified common regions across the populations, where genes related to a variety of traits were reported, such as body size, muscle development, and brain functions. These results indicate that such regions are associated with many traits, and thus were under selective pressure in sheep breeds raised for different purposes. Interestingly, many candidate genes detected within the HRR islands were associated with brain integrity. We also observed a strong association of high linkage disequilibrium pattern with ROH compared with HRR, despite the fact that many regions in linkage disequilibrium were not located in ROH regions.

Some sectors of animal production and reproduction have shown great technological advances due to the development of research areas such as Precision Livestock Farming (PLF). PLF is an innovative approach that allows animals to be monitored, through the adoption of cutting-edge technologies that continuously collect real-time data by combining the use of sensors with advanced algorithms to provide decision tools for farmers. Artificial Intelligence (AI) is a field that merges computer science and large datasets to create expert systems that are able to generate predictions and classifications similarly to human intelligence. In a simplified manner, Machine Learning (ML) is a branch of AI, and can be considered as a broader field that encompasses Deep Learning (DL, a Neural Network formed by at least three layers), generating a hierarchy of subsets formed by AI, ML and DL, respectively. Both ML and DL provide innovative methods for analyzing data, especially beneficial for large datasets commonly found in livestock-related activities. These approaches enable the extraction of valuable insights to address issues related to behavior, health, reproduction, production, and the environment, facilitating informed decision-making. In order to create the referred technologies, studies generally go through five steps involving data processing: acquisition, transferring, storage, analysis and delivery of results. Although the data collection and analysis steps are usually thoroughly reported by the scientific community, a good execution of each step is essential to achieve good and credible results, which impacts the degree of acceptance of the proposed technologies in real life practical circumstances. In this context, the present work aims to describe an overview of the current implementations of ML/DL in livestock reproduction and production, as well to identify potential challenges and critical points in each of the five steps mentioned, which can affect results and application of AI techniques by farmers in practical situations.

Background: A cost-effective strategy to explore the complete DNA sequence in animals for genetic evaluation purposes is to sequence key ancestors of a population, followed by imputation mechanisms to infer marker genotypes that were not originally reported in a target population of animals genotyped with single nucleotide polymorphism (SNP) panels. The feasibility of this process relies on the accuracy of the genotype imputation in that population, particularly for potential causal mutations which may be at low frequency and either within genes or regulatory regions. The objective of the present study was to investigate the imputation accuracy to the sequence level in a Nellore beef cattle population, including that for variants in annotation classes which are more likely to be functional. Methods: Information of 151 key sequenced Nellore sires were used to assess the imputation accuracy from bovine HD BeadChip SNP (~ 777 k) to whole-genome sequence. The choice of the sires aimed at optimizing the imputation accuracy of a genotypic database, comprised of about 10,000 genotyped Nellore animals. Genotype imputation was performed using two computational approaches: FImpute3 and Minimac4 (after using Eagle for phasing). The accuracy of the imputation was evaluated using a fivefold cross-validation scheme and measured by the squared correlation between observed and imputed genotypes, calculated by individual and by SNP. SNPs were classified into a range of annotations, and the accuracy of imputation within each annotation classification was also evaluated. Results: High average imputation accuracies per animal were achieved using both FImpute3 (0.94) and Minimac4 (0.95). On average, common variants (minor allele frequency (MAF) > 0.03) were more accurately imputed by Minimac4 and low-frequency variants (MAF ≤ 0.03) were more accurately imputed by FImpute3. The inherent Minimac4 Rsq imputation quality statistic appears to be a good indicator of the empirical Minimac4 imputation accuracy. Both software provided high average SNP-wise imputation accuracy for all classes of biological annotations. Conclusions: Our results indicate that imputation to whole-genome sequence is feasible in Nellore beef cattle since high imputation accuracies per individual are expected. SNP-wise imputation accuracy is software-dependent, especially for rare variants. The accuracy of imputation appears to be relatively independent of annotation classification.

The world population is expected to reach 9.1 billion people by 2050 (FAO, 2019) and the largest area of populational growth is projected to occur in developing countries. Consequently, there is an urgent need to increase food production in these regions based on an efficient exploitation of a wide range of genetic resources (species, breeds) and production systems. In this context, South America is one of the major livestock and aquaculture producers in the world, especially beef, pork, poultry, South American camelids, tilapia, and salmon. As in other developing regions, the great productivity is accompanied by a large diversity of production systems (e.g., intensive, pasture-based, mountainous regions, and small holders), climatic and geographical conditions, availability of natural resources (e.g., water and forage), and genetic resources (numerous species, breeds, and populations). In order to meet the growing food demand and be competitive on a global scale, there is a need to increase production efficiency and therefore, long-term sustainability. This can be achieved through selective breeding by identifying animals with greater genetic merit for the traits of interest. Sustainable genetic improvement can only be achieved through accurate, routine recording, and proper use of at least phenotypic and pedigree information.

This study focused on assessing the usefulness of using audio signal processing in the gaited horse industry. A total of 196 short-time audio files (4 s) were collected from video recordings of Brazilian gaited horses. These files were converted into waveform signals (196 samples by 80,000 columns) and divided into training (N = 164) and validation (N = 32) datasets. Twelve single-valued audio features were initially extracted to summarize the training data according to the gait patterns (Marcha Batida—MB and Marcha Picada—MP). After preliminary analyses, high-dimensional arrays of the Mel Frequency Cepstral Coefficients (MFCC), Onset Strength (OS), and Tempogram (TEMP) were extracted and used as input information in the classification algorithms. A principal component analysis (PCA) was performed using the 12 single-valued features set and each audio-feature dataset—AFD (MFCC, OS, and TEMP) for prior data visualization. Machine learning (random forest, RF; support vector machine, SVM) and deep learning (multilayer perceptron neural networks, MLP; convolution neural networks, CNN) algorithms were used to classify the gait types. A five-fold cross-validation scheme with 10 repetitions was employed for assessing the models' predictive performance. The classification performance across models and AFD was also validated with independent observations. The models and AFD were compared based on the classification accuracy (ACC), specificity (SPEC), sensitivity (SEN), and area under the curve (AUC). In the logistic regression analysis, five out of the 12 audio features extracted were significant (p < 0.05) between the gait types. ACC averages ranged from 0.806 to 0.932 for MFCC, from 0.758 to 0.948 for OS and, from 0.936 to 0.968 for TEMP. Overall, the TEMP dataset provided the best classification accuracies for all models. The most suitable method for audio-based horse gait pattern classification was CNN. Both cross and independent validation schemes confirmed that high values of ACC, SPEC, SEN, and AUC are expected for yet-to-be-observed labels, except for MFCC-based models, in which clear overfitting was observed. Using audio-generated data for describing gait phenotypes in Brazilian horses is a promising approach, as the two gait patterns were correctly distinguished. The highest classification performance was achieved by combining CNN and the rhythmic-descriptive AFD.